The dietary glycaemic index (GI) of a carbohydrate-containing food is an indicator of its effect on postprandial blood glucose response, compared with the response elicited by administration of glucose with equivalent carbohydrate content. Epidemiological studies suggest a possible role of dietary GI in increasing the risk of cancer at some sites( Reference Melkonian, Daniel and Ye 1 – Reference Sieri, Krogh and Agnoli 3 ), and there appears to be a biological plausibility for such an association( Reference McKeown-Eyssen 4 – Reference Aranceta Bartrina and Perez Rodrigo 6 ). GI has been found to modulate insulin sensitivity( Reference Jenkins, Wolever and Taylor 7 ). Low-GI diets compared with high-GI diets may improve glycaemic control, leading to lower insulin output and inflammatory responses and possibly a lower risk of several chronic diseases, including cancer( Reference Mullie, Koechlin and Boniol 2 , Reference Barclay, Petocz and McMillan-Price 8 , Reference Choi, Giovannucci and Lee 9 ). GI varies considerably for different foods depending on several factors such as the amounts and types of carbohydrates and the cooking method. The product of a food’s GI and its total carbohydrate content is represented by another indicator called the glycaemic load (GL). GL accounts not only for carbohydrate quality but also for quantity( Reference Foster-Powell, Holt and Brand-Miller 10 ). Findings from the Italian component (European Prospective Investigation into Cancer and Nutrition (EPIC)-Italy) of the EPIC initiative have suggested that GL and carbohydrate intake are positively associated with breast cancer and colon cancer risk( Reference Sieri, Krogh and Agnoli 3 , Reference Romieu, Ferrari and Rinaldi 11 ).
The risk of malignant melanoma, which has been associated with ultraviolet exposure, atypical nevi and genetic factors, has also been found to be linked to dietary factors( Reference Murzaku, Bronsnick and Rao 12 – Reference Miura and Green 14 ). In a previous study, we found an association between carbohydrate intake and melanoma risk( Reference Vinceti, Pellacani and Malagoli 15 ), and insulin resistance has been suggested to occur in melanoma patients( Reference Antoniadis, Petridou and Antonopoulos 16 ). However, no study appears to have evaluated the association between GI and GL and risk of melanoma. We sought to address this question using a population-based, case–control study in an Italian community.
Methods
Details of our case–control study on dietary risk factors for cutaneous melanoma in the population of five provinces of the Emilia Romagna region, northern Italy (population about 3 000 000), have been provided elsewhere( Reference Vinceti, Malagoli and Iacuzio 17 – Reference Malagoli, Malavolti and Agnoli 19 ). In brief, during 2005–2006, we attempted to recruit all patients with newly diagnosed cutaneous malignant melanoma residing in the provinces of Bologna, Ferrara, Modena, Parma and Reggio Emilia and attending the local dermatological clinics. Inclusion criteria were a histologically based recent diagnosis of cutaneous melanoma without clinical evidence of metastasis. A total of 572 eligible patients were contacted by their dermatologists to participate in the study, and 394 (68·9 %) agreed to participate. In these subjects, the melanomas had a median Breslow’s thickness of 0·61 mm, with no substantial differences between males and females (0·179 by non-parametric K-sample test on the equality of medians). All patients had undergone only surgical treatment for the disease. Each subject was administered lifestyle and food frequency questionnaire (FFQ) (see below) during a routine visit at the beginning of follow-up for their disease. The questionnaires could be returned at the following visit or sent back to the Research Center for Environmental, Genetic and Nutritional Epidemiology of Modena and Reggio Emilia University. We randomly selected 2825 potential controls matched to cases for sex, year of birth (±5 years) and province of residence from the Emilia Romagna region National Health Service directory (mandatory for all Italian residents), and we mailed lifestyle and FFQ, a description of the study and a pre-paid return envelope to them. A total of 747 (26·4 %) potential controls agreed to participate in the present study and returned the questionnaires. In all, fourteen cases and twenty-eight controls were excluded from subsequent analysis because of incompleteness of data or extreme values (ratio of total energy intake:calculated basal metabolic rate lower than the 0·5th percentile or higher than the 99·5th percentile) derived from the FFQ. All participants gave their written informed consent before enrolment.
Dietary assessment
We investigated subjects’ usual diet during the past 12 months with a validated, semi-quantitative FFQ designed to capture eating behaviours in Italy, specifically developed as part of the EPIC study for the northern Italian population( Reference Pala, Sieri and Palli 20 , Reference Salvini, Parpinel and Gnagnarella 21 ). The EPIC questionnaire was designed to be self-administered, and was checked by trained personnel after compilation. Participants were asked to respond to 248 questions about 188 different food items, including seasonal foodstuffs, and to indicate the number of times a given item was consumed (per day, week, month or year), from which the absolute frequency of consumption of each item was calculated. The quantity of food consumed was assessed by selecting an image of a food portion or by selecting a predefined standard portion when no image was available. The food items were then linked, using specially designed software( Reference Pala, Sieri and Palli 20 ), to the Italian Food Tables( Reference Salvini, Parpinel and Gnagnarella 21 ) to obtain estimates of daily intake of macronutrients and micronutrients plus energy. GI of food items containing available carbohydrates were obtained from the Italian Glycemic Index Table (managed by the Department of Food Sciences, University of Parma, based on direct analysis of main carbohydrate food sources in Italy), which contains about 150 food products covering >90 % of the carbohydrate intake of people living in Italy. When a food consumed was not present in the table, we used the GI value published elsewhere (GI international tables( Reference Atkinson, Foster-Powell and Brand-Miller 22 ) and www.glycemicindex.com ( Reference Sieri, Krogh and Berrino 23 )). We computed the daily average dietary GI for each participant as the sum of the GI of each food item consumed, multiplied by the average daily amount consumed and the percentage of carbohydrate content, all divided by the total daily carbohydrate intake. The daily average dietary GL was calculated similarly except that there was no division by total carbohydrate intake. Total carbohydrate intake (g/d) was calculated using the Food Composition Database for Epidemiologic Research in Italy published by the European Oncology Institute( 24 ).
Additional variables
Each participant provided information on demographic characteristics (place and date of birth, province of residence, marital status, education), weight and height, phenotypic characteristics (eye, hair and skin colour), sunburn history (never, first before 18 years of age, first after 18 years of age) and skin reaction to sunlight exposure (speed of tan and tendency to burn). We classified eye colour into light (blue/green), light brown and dark (brown/black) categories. Hair colour was classified as blond, red, light brown or dark brown/black at 20 years, and skin colour was classified as white, light brown, brown/olive or dark brown/ebony. On the basis of these categories, each subject was assigned to a phototype using the Fitzpatrick phototyping scale( Reference Fitzpatrick 25 ). We computed BMI as weight/height2 (kg/m2) and categorised subjects as normal weight, overweight or obese according to commonly used definitions( 26 ).
Statistical analysis
Analyses were conducted overall and stratified by sex. We computed the odds ratio (OR) for melanoma according to quintiles of distribution (among controls) of GI, GL and total carbohydrates, sugars and starch after adjusting for energy intake using the Willett nutrient residual method( Reference Willett and Stampfer 27 ). OR and 95 % CI were computed using conditional logistic regression analysis, adjusting for education (four categories), BMI (continuous), phototype (four categories), skin sensitivity to sun exposure (five categories), sunburns history (three categories) and dietary intake of total energy (continuous), vitamin C (continuous), vitamin D (continuous), saturated fatty acids (SFA) (continuous) and fibre (continuous). Trends in the associations between GI and GL scores and risk were assessed by computing P values based on their values as continuous variables in conditional logistic regression models( Reference Rothman, Greenland and Lash 28 ). Finally, we modelled the association between GI and GL scores and risk of cutaneous melanoma using restricted cubic splines, computed with mkspline and xblc commands of Stata 14 statistical software (StataCorp LP, 2015)( Reference Orsini and Greenland 29 ). We selected the optimal number of knots according to Akaike’s information criterion and used the knot-placement method recommended by Harrell( Reference Harrell 30 ), which led us to place three knots, at the 10th, 50th and 90th percentiles.
Results
A total of 380 cases (175 men and 205 women, aged 58±16 and 53±15 years, respectively) and 719 age-, sex- and province of residence-matched controls were included in the analysis. Demographic and clinical characteristics of cases and controls are reported in Table 1. Educational attainment and marital status were similar in cases and controls, whereas cases tended to have more fair skin types, high tendency to burn and were more likely to report a history of sunburns, ever and particularly after 18 years. Dietary characteristics related to the amount of carbohydrates, summarised in Table 2, showed that cases tended to have slightly higher GI and slightly lower total sugar intakes compared with controls and a higher intakes of starch. GI and GL were positively correlated (r 0·44; 95 % CI 0·39, 0·48).
Table 1 Characteristics of the study participants by case–control status (Numbers and percentages)
* P values from χ 2 tests.
† Phototype I, eyes/hair/skin light, high tendency to burn and never/moderate tan; phototype II, eyes/hair/skin light, moderate tendency to burn and gradual tan, or eyes/hair/skin brown high tendency to burn and moderate tan; phototype III, eyes/hair/skin brown, moderate/no tendency to burn and gradual/golden tan; phototype IV, no tendency to burn and intense tan.
Table 2 Dietary characteristics of the study participants by case–control status (Mean values and standard deviations)
* Daily dietary average.
† P values from two sample t tests.
Table 3 presents the unadjusted and adjusted OR for melanoma and corresponding 95 % CI according to quintiles of GI, GL and other dietary carbohydrates for the overall sample. The results suggest a positive association between GL and melanoma risk with OR 1·53 (95 % CI 1·02, 2·30) after comparison between the highest (Q5) and the lowest (Q1) quintiles in the unadjusted model, which was attenuated and less precise in the adjusted model (OR 1·35; 95 % CI 0·80, 2·27, Q5 v. Q1). No positive association emerged between GI and melanoma risk. The results also suggest that a higher risk of melanoma is associated with higher intakes of starch (adjusted OR 1·88; 95 % CI 1·05, 3·40, Q5 v. Q1).
Table 3 Overall OR for cutaneous melanoma according to quintiles of daily average of glycaemic load, glycaemic index and total carbohydrates, total sugars and starch intakesFootnote † (Odds ratios and 95% confidence intervals; cases/controls, medians)
* P for linear trend based on continuous values of intake.
† Lowest quintile (Q1) as the referent category.
‡ Unadjusted analysis.
§ Adjusted for intakes of SFA, vitamin C, vitamin D, fibre and total energy, phototype, skin sensitivity to sun exposure, sunburns history, BMI and education.
Table 4 summarises analyses stratified by sex, adjusting for all covariates or only for energy and two dietary factors previously found to be associated with disease risk in our study population, vitamins C and D. The association between GL and melanoma risk was observed among females only (fully adjusted OR 2·40; 95 % CI 1·23, 4·70, Q5 v. Q1), although the OR across the GL quintiles did not increase linearly. No evidence of any association with GI emerged in either sex. Total carbohydrate intake was also associated with melanoma risk only in females (fully adjusted OR 2·18; 95 % CI 1·16, 4·10, Q5 v. Q1), whereas the inverse association with sugar intake was observed only in men (OR 0·37; 95 % CI 0·17, 0·80, Q5 v. Q1). There was suggestion of a positive association of melanoma risk with starch intake for men (OR 1·99; 95 % CI 1·04, 3·80, Q5 v. Q1) and for women (OR 2·00; 95 % CI 1·11, 3·61, Q5 v. Q1), although these OR were attenuated and less precise in fully adjusted analyses. Stratified analyses for BMI, age and phototype did not reveal differences (data not shown).
Table 4 Adjusted OR for melanoma in sex-stratified analysisFootnote † (Adjusted odds ratios and 95 % confidence intervals)
* P for linear trend based on continuous values of intake.
† Lowest quintile (Q1) as the referent.
‡ Adjusted for vitamin C, vitamin D and total energy intakes.
§ Adjusted for intakes of SFA, vitamin C, vitamin D, fibre and total energy, phototype, skin sensitivity to sun exposure, sunburns history, BMI and education.
Plots estimating the association between odds of being a case and GI and GL values, produced using regression spline analysis, appeared to confirm a direct association between GL and melanoma risk in females (Fig. 1).
Fig. 1 Restricted cubic spline regression analysis of the odds of being a case according to the glycaemic load score, adjusting for SFA, vitamin C, vitamin D, fibre and total energy intakes, phototype, skin sensitivity to sun exposure, sunburns history, BMI and education. 
, 95 % confidence limits. Reference line at 1·0.
Discussion
Although cutaneous melanoma has generally not been attributed to lifestyle factors apart from recreational or occupational ultraviolet exposure, our study is in line with growing epidemiological evidence suggesting a role of dietary factors in disease aetiology( Reference Murzaku, Bronsnick and Rao 12 – Reference Vinceti, Pellacani and Malagoli 15 , Reference Vinceti, Malagoli and Iacuzio 17 – Reference Malagoli, Malavolti and Agnoli 19 , Reference Vinceti, Bonvicini and Pellacani 31 – Reference Loftfield, Freedman and Graubard 34 ). In the present investigation, apparently the first to evaluate an association of melanoma with GI and GL, we identified an association of GL with disease risk, although only in females and not always with a consistent trend. The observation of an association between dietary factors and melanoma risk only in females is not a new finding, as it has already been observed in all previous studies carried out in this study population( Reference Vinceti, Pellacani and Malagoli 15 , Reference Malagoli, Malavolti and Agnoli 19 , Reference Malavolti, Malagoli and Fiorentini 32 ) except one( Reference Vinceti, Malagoli and Fiorentini 18 ). In interpreting these results, it must be considered that GL but not GI takes into account the amount of carbohydrate consumed in addition to qualitative factors such the concomitant intake of carbohydrates, fats and proteins, the processing of foods, meal preparation and serving temperature. Therefore, dietary GL is a more sensitive measure of postprandial glycaemia and insulin demand than dietary GI( Reference Bao, Atkinson and Petocz 35 , Reference Roberts and Liu 36 ). If melanoma risk is related to the overall insulin demand, it could be expected to be more strongly related to dietary GL than to dietary GI, consistent with what we observed. Our findings can be interpreted in light of previous studies that have found an association between GL, GI and risk of different cancers( Reference Mullie, Koechlin and Boniol 2 , Reference Sieri, Krogh and Agnoli 3 , Reference Turati, Galeone and Gandini 37 , Reference Woo, Park and Shin 38 ). Some studies have found the association with disease risk to be stronger for, or exclusive to, GL compared with GI( Reference Oba, Nanri and Kurotani 39 – Reference Arikawa, Jakits and Flood 41 ).
The pathophysiological mechanism that would link GI or GL to the aetiology of melanoma is still unknown. Gogas et al.( Reference Gogas, Trakatelli and Dessypris 42 ) suggested a possible role of leptin in melanoma development. Leptin levels are directly associated with obesity, insulin levels and female sex( Reference Havel, Kasim-Karakas and Dubuc 43 , Reference Thomas, Burguera and Melton 44 ), which may explain our findings of an increased risk of melanoma in females only. This sex-specific effect was also not unexpected on the basis of previous results in this study sample, as diet quality and vitamin C intake had shown an inverse association with disease risk that was largely restricted to females( Reference Malagoli, Malavolti and Agnoli 19 , Reference Malavolti, Malagoli and Fiorentini 32 ), already suggesting a potential effect modification by sex.
There is growing evidence supporting an association of dietary indices such as GL and GI with cancer risk at some sites( Reference Melkonian, Daniel and Ye 1 , Reference Mullie, Koechlin and Boniol 2 , Reference Turati, Galeone and Gandini 37 , Reference Qin, Moorman and Alberg 45 , Reference Amadou, Degoul and Hainaut 46 ). The observation of a possible association of the disease with metabolic syndrome and altered insulin sensitivity( Reference Antoniadis, Petridou and Antonopoulos 16 ) and, in our study population, of an inverse association between soluble carbohydrate intake and melanoma risk, particularly in females( Reference Vinceti, Pellacani and Malagoli 15 ), add some plausibility to an association of GI and/or GL with melanoma risk.
Our study has certain limitations. The FFQ was not originally designed to estimate dietary GI and GL, although it carefully assessed total carbohydrate and energy intakes. In particular, GI and GL estimated through the EPIC FFQ may not accurately reflect the glycaemic effects of consuming mixed dishes as compared with individual food items. Another study limitation is the reduced statistical precision of the risk estimates for some subgroups, as shown by the wide CI. We also recognised that the study findings should be further assessed in studies with cohort designs to overcome the general limitations of case–control investigations.
The study also has strengths. Both selection and recall bias are unlikely, as diet was not generally regarded as a risk factor for melanoma, and we controlled for many potential confounders in the analyses. We also note that in our population the GI score distribution was comparable with that observed in other surveys conducted in the Italian population, whereas the GL score distribution was slightly lower( Reference Sieri, Brighenti and Agnoli 47 ).
In conclusion, we observed an association between GL and melanoma risk in an Italian population that was limited to females, although the possibility of unmeasured confounding cannot be ruled out. Further studies are warranted to fully elucidate the role of GL and related dietary factors in predicting risk of melanoma in humans, as well as to identify the reasons limiting this association to females.
Acknowledgements
Financial support for this study was provided by the Local Health Unit of Reggio Emilia – Italy (M. V.), the Italian League against Cancer – LILT Reggio Emilia Section (M. V.), Modena Policlinico Hospital and Modena Local Health Unit (M. V. and G. P.), and the US National Institute of Health (C. M. C., grant no. NIH-CA16042). All these funders had no role in the design, analysis or writing of this article.
M. V., C. M. and G. P. designed the original study; M. M., C. M. C., F. B., S. S. and C. A. analysed and interpreted the data and drafted the article. C. M. and M. V. recruited controls and collected their data. F. B., C. A., S. S. and V. K. prepared the FFQ and the associated nutrient and energy database and computed the glycaemic index of food items. C. F., F. F., C. L., C. R., G. A., A. L., L. V., A. V., C. P., C. F., P. A. F., E. D. and G. P. enrolled melanoma patients and collected their clinical, lifestyle and dietary data. All the authors read and approved the final version of the manuscript.
The authors declare that there are no conflicts of interest.
